Le jeudi 09 février 2023 à 18:54 +0000, John Cox a écrit : > > Maybe insert a phrase related to the use of the same columns pattern as > > NV12_C128 ? Then from there, its easier to understand why this bytesperline > > formula. I'm guessing that 96 is the number of pixels that fits 1 row of 128 > > bytes (128 / 4 * 3). I could not guess why you need 128 times that size > > though ? > > Maybe I'm missing something ? > > 96 is indeed the number of pixels that fits in one 128 byte column. > > (width+95)/96 is the number of columns required, mutiplying it by 128 > gets you a "pitch" that when mutiplied by height gives you the number of > bytes occupied by the luma plane. This is mostly to keep software that > expects width <= bytesperline happy. Thanks, now that you say it, its kind of obvious, its pixel to byte conversion. And I agree this is probably the best emulation we can do, and documenting it is important. Notion of stride is also a bit fuzzy with other tile formats. And I suspect this convention is also that one use on DRM side ? Nicolas
